Methodology

How the ValPro model arrives at a value

The model is a hedonic regression: it prices the individual characteristics of homes in a submarket, then applies those prices to the subject property. The same discipline an appraiser applies through paired-sales analysis, executed across every available sale.

1. Subject identification

The address is normalized and matched to the public records. We retrieve all of the physical characteristics of the subject property, including but not limited to location, living area size, room counts, age, garage information, and more.

2. Comparable sales retrieval

We query recorded residential sales in the subject's submarket over the trailing 15 months, filtered to a living-area band around the subject and a radius that keeps the sample within the same market area. A typical search returns 35–50 usable transactions.

3. Hedonic regression

A multiple linear regression is fit by least squares on the retrieved sample, with sale price as the dependent variable and living area, bedrooms, bathrooms, lot size, effective age, garage spaces, and months since sale as independent variables. A small ridge penalty stabilizes coefficients when predictors are collinear (for example living area and bedroom count). This is precisely where our extensive traditional appraisal experience starts to separate ValPro from outdated and less accurate AVMs of the past.

4. Similarity scoring

Every sale is scored 0–100 against the subject using scale-normalized gaps in living area, bedrooms, bathrooms, lot size, year built, distance, and recency. The four highest scores become the reported comparables.

5. Adjustment and reconciliation

Each comparable's sale price is adjusted to the subject's characteristics using the fitted coefficients. The final indicated value is a weighted reconciliation: 60% of the direct regression prediction and 40% of the similarity-weighted adjusted comparable prices.

6. Diagnostics and range

We report adjusted R², the residual standard error, and the sample size. The value range is the wider of one standard error or ±4% of the indicated value. Confidence blends explanatory power, comparable quality, and sample depth.

Regression specification

Price = β0
      + β1·LivingArea      (sqft)
      + β2·Bedrooms
      + β3·Bathrooms
      + β4·LotSize         (sqft)
      + β5·EffectiveAge    (years)
      + β6·GarageSpaces
      + β7·MonthsSinceSale
      + ε

β is estimated by solving the ridge-regularized normal equations (XᵗX + λI)β = Xᵗy with Gaussian elimination and partial pivoting.